What I'm working on

• Modular Algorithmic Trading Framework

  The aim of this project is to have a system that allows to speed up the screening phase where multiple strategies are evaluated to pass to the next one where the most robust ones are implemented and optimized.

  
  

  It provides a framework to implement fully automatic trading strategies on Forex market for backtesting and live trading. The scripts use the MetaTrader5 Python API to connect with the broker and execute the orders. It's also available a feature to receive instant messages from a Telegram Bot whenever an order is executed or a position is closed, showing the profit of the latter.

  
  

  The entire system is built to be modular, so different types of strategies can be implemented with the same approach and with little to no changing in the main code. Some brief examples of strategies are statistical and econometrics-based as well as ML strategies using neural networks or other approaches.

• Optimization - Extreme Learning Machine

  This project aims to evaluate the performances of gradient based methods in the learning process of an Extreme Learning Machine (ELM) that typically makes use of a single step approach - the computation of the Moore-Penrose pseudo-inverse matrix.

  
  

  Looking at the problem from an optimization point of view, the considered function is convex and non-differentiable; two algorithms are tested, the momentum gradient descent method (that tipically fails on non-differentiable functions) and the deflected subgradient method. Tests will be executed on some well-known datasets (MNIST, CIFAR10, CIFAR100, CoverType).